Adversarial Policies: Attacking Deep Reinforcement Learning
This addresses security vulnerabilities in RL systems for applications like robotics, though it is incremental as it builds on known adversarial example concepts.
The paper tackles the problem of attacking deep reinforcement learning agents in multi-agent environments by training adversarial policies that create natural but adversarial observations, demonstrating that these policies reliably defeat state-of-the-art victims trained via self-play in simulated humanoid robot games.
Deep reinforcement learning (RL) policies are known to be vulnerable to adversarial perturbations to their observations, similar to adversarial examples for classifiers. However, an attacker is not usually able to directly modify another agent's observations. This might lead one to wonder: is it possible to attack an RL agent simply by choosing an adversarial policy acting in a multi-agent environment so as to create natural observations that are adversarial? We demonstrate the existence of adversarial policies in zero-sum games between simulated humanoid robots with proprioceptive observations, against state-of-the-art victims trained via self-play to be robust to opponents. The adversarial policies reliably win against the victims but generate seemingly random and uncoordinated behavior. We find that these policies are more successful in high-dimensional environments, and induce substantially different activations in the victim policy network than when the victim plays against a normal opponent. Videos are available at https://adversarialpolicies.github.io/.